Re: [PATCH 20/29] crypto: talitos - Replace SEC1/SEC2 conditionals with ops dispatch

[Christophe Leroy (CS GROUP)]

Le 04/06/2026 à 15:05, Paul Louvel a écrit :
> On Thu Jun 4, 2026 at 11:37 AM CEST, Christophe Leroy (CS GROUP) wrote:
> > Le 28/05/2026 à 11:08, Paul Louvel a écrit :
> > > Replace the if/else is_sec1 dispatches in callers with indirect calls
> > > through priv->ops. Add static const sec1_ops and sec2_ops structs
> > > populated with the SEC1 and SEC2 function variants, and set priv->ops
> > > at probe time based on the detected hardware.
> >
> > Why is that needed ?
> >
> > I understand your objective at the end is to get rid of that is_sec1
> > boolean that is carried over the entire call chain but using ops for
> > that seems overkill.
> >
> > What about changing it to a helper using static branches, something like
> > (untested) :
> >
> > #if defined(CONFIG_CRYPTO_DEV_TALITOS1) &&
> > defined(CONFIG_CRYPTO_DEV_TALITOS2)
> > DECLARE_STATIC_KEY_FALSE(talitos_is_sec1);
> > static __always_inline bool is_sec1(void)
> > {
> > 	return static_branch_unlikely(&talitos_is_sec1);
> > }
> >
> > static inline void talitos_init_branch(bool is_sec1)
> > {
> > 	if (is_sec1)
> > 		static_branch_enable(&talitos_is_sec1);
> > }
> > #else
> > static __always_inline bool is_sec1(void)
> > {
> > 	return IS_ENABLED(CONFIG_CRYPTO_DEV_TALITOS1);
> > }
> >
> > static inline void talitos_init_branch(bool is_sec1)
> > {
> > 	BUILD_BUG_ON(is_sec1 && !IS_ENABLED(CONFIG_CRYPTO_DEV_TALITOS1));
> > }
> > #endif
>
> Thanks you for that suggestion.
> This was a lack of knowledge about this mechanism.

static_branch is nice for small inlined helpers.

For bigger time critical functions, you have static calls.

See exemple in commit e59596a2d6a7 ("powerpc: Use static call for 
get_irq()")

This has become even more efficient since commit f50b45626e05 
("powerpc/static_call: Implement inline static calls")

Christophe